Skip to main contentCúram Performance & Tuning Guide

Adjusting thread tuning

If the system throughput is known or a good estimation of the system throughput is available, you can calculate the number of threads that are in the web server. You can do the calculation after the persistent connections are configured.

The following formula indicates the total number of threads that are required in the web server:

Throughput * KeepAliveTimeOut

Alternatively, you can use the following formula to obtain a rough starting value for the number of threads that you can use, where WebContainer_max is the number of application server threads that are available to process HTTP requests:

WebContainer_max * KeepAliveTimeOut

If the HTTP server load balances against multiple application servers, multiply the number of threads as needed:

ratio_of_application_servers_per_HTTP_servers * WebContainer_max * KeepAliveTimeOut

The previous thread calculation is likely to be too low a number, but you can use it as a quick start. Then, refine the number of threads during load tests or production monitoring by including the 95th percentile of response time through the following heuristic:

threads = ratio_of_application_servers_per_HTTP_servers * WebContainer_max *
(KeepAliveTimeOut + user_interaction_95th%ile_response_time)

You might need to adjust the number of threads further depending on the following factors:

  • The average amount of static content that is in your web pages
  • How well the static content is cached by the web browser

The dependency occurs because some web clients can use multiple connections for retrieving the static content on a web page. The behavior is a strong case for monitoring the HTTP server threads utilization.

During monitoring, as a general convention, increase the number of threads only if the following statements are true:

  • All threads are active, and no idle threads exists.
  • The response time is not acceptable.
  • The CPU is not fully utilized.

The procedure for setting the number of threads differs according to the operating system:

ServerLimit           1
ThreadLimit           300
StartServers          1
MaxClients            300
MinSpareThreads       300
MaxSpareThreads       300
ThreadsPerChild       300
MaxRequestsPerChild   0
Previous
Tuning the HTTP server: Persistent Connections
Next
Tuning on VMs: Tuning the application server